Thank you for the long form video, walking through the code and explaining the caveats so clearly. We have missed you. Hopefully there will be more content coming our way soon. Cheers and stay safe.
Brilliant video. New to the Nano. Gleaned ALOT from the vid. Never thought to thread the camera capture, was a "I could have had a V8." moment. Project I'm working on mandates high frame rates, now I got a clue. Big thanks.
Thank you for the kind words. Threading is usually the first trick to use on smaller machines if available. The goal is to try to take advantage of as much of the machine resources as possible. The smaller the machine, the more you have to know to make it performant. Thanks for watching!
@@JetsonHacks Yeah, concur regarding the need to get closer to the hardware & do a little "bit twiddling". I'm an "old school" coder. Cut my teeth on Univac main frames in the 70's-80's, transitioned to the 8086, 6800, & 68000 families in the late 80's, quite coding professionally in the middle to late 90's. Worked w/ the Nestor NI1000, back in the day (1,024 neurons in silicon) . My forte, language wise, was machine language, assembly (on several chips sets), C, C++, Ada, & several high level languages so Python is new to me, it being a relatively new OO language. So your "tutorials" are beneficial in multiple arenas. My dev machine is a HP DL580 (40 cores/80 threads, 1Tb memory, 4x 1080Ti's, etc.), am amazed @ just how much AI @ the edge can do (RPi 4 + NCS, Nano, AGX). Ever done anything combining the Nano + NCS (OpenCV + OpenVINO + CuDNN)? Think it would be interesting to encapsulate NLP "at the edge" versus the Alexa/Wolfram Alpha client server style implementation. Anyway, keep up the good teaching/experimentation, I'll be watching. What's the best way to subsidize your endeavors here on TH-cam, et al.? Feel the least I can do is throw you some pizza $$$. :)
I didnt order the Jetson Nano last year :) - but i came back to Your channel to check up if there's something new coming out :) - Thank You for a new video :)
THIS IS AWESOME! thank you for this primer, next i need to feed each frame into a real time SLAM algorithm and build a volumetric representation of the space around the drone so it can A* path hmm
I am currently working on a project using an inland camera and when I try to test the camera this error shows up: "Error generated. /dvs/git/dirty/git-master_linux/multimedia/nvgstreamer/gst-nvarguscamera/gstnvarguscamerasrc.cpp, execute:557 No cameras available" can anyone help?
Imo, firstly it's better to check if everything is ok with cameras using ls /dev/video* You can see 2 cameras with video0 and video1 names, means both must visible as devices. otherwise gst gives errors, wich are difficult to understand. Also dual camera sample didn't work for me with sensor_mode parameter, I've commented them out, then it was ok.
Hi! I purchased the cam too. For stereo vision with Jetson Nano, could you help me? I have tried to solve the connection between the cam and Jetson Nano.
i am currently using it, it works the same maybe try changing CSI settings to IMX219 from jetson-io (th-cam.com/video/eImDQ0PVu2Y/w-d-xo.html around 9-min guy uses the same program to configure pwm pins)
There are tradeoffs. Typically stereo cameras have global shutter and hardware sync which using two RPi cameras do not offer. The results won't be perfect, but you should be able to play around with the concept. Thanks for watching!
Hello Jim, i took a tutorial on your videos and did everything. But some point i got stuck with the syncing of two cameras in instrumened folder. I got an error like this: python -u "/home/jetson/Documents/3D_workspace/CSI-Camera/instrumented/dual_camera_fps.py" Traceback (most recent call last): File "/home/jetson/Documents/3D_workspace/CSI-Camera/instrumented/dual_camera_fps.py", line 16, in from csi_camera import CSI_Camera File "/home/jetson/Documents/3D_workspace/CSI-Camera/instrumented/csi_camera.py", line 14, in class RepeatTimer(threading.Timer): TypeError: Error when calling the metaclass bases function() argument 1 must be code, not str Can you let me know what the error is.. Thanks in advance.
Hi ... thanks for the wonderful tutorials .. by the way, i just would like to know if you had experienced connecting flir lepton breakout board on jetson nano?
Hi, I could only get a hold on a jetson nano A02 here in our country.. Is it possible to use 2 usb cameras in a stereo setup for depth estimation using A02?
You make it look so easy, but it's really not, at least for me. Looks like I'm going to have to devote a lot more time to this. There must be a way to do facial recognition from off axis. There's always a way to do something that seems imposible. I like that the fire extinguisher is always in frame. You should plug in your guitar and crank out some power cords at the end of some of your videos. I like saying rock on dude, instead of stay safe. I say that all the time.
Why not do a while loop around the cv2.waitKey checking the internal clock until 1000/fps has elapsed since you last presented a frame, and sleep for like, 1ms so it sleeps for exactly the right amount of time to match the frame rate?
Are there any differences in the board size between a02 and b01 ? Board screw holes etc? Tried finding the cad files for both, but didn't see them (still looking btw) :)
Hi there, wonderful video.. I have tried everything you said but still getting errors after running the Python code in visual studio.. not sure why.. I have installed everything, cython, numpy
@@JetsonHackswhen running the Python codes like face detect, simple camera scripts, it's throwing number of errors and I can't see the camera window. No matter what I do. Not sure if I am missing any plugins or software . I did whatever you have mentioned in your video
Hello, Can you please tell what is the maximum fps that can be achievable using Jetson nano and RPi camera(dual as well as single)? In general a RP camera specfi max is written as 120/90fps but I guess using OpenCV or bash command we can achieve higher. here using dual camera along with Jetson nano what's the max FRAME PER SEC limit is possible?
Please ask your question on the official NVIDIA jetson Nano forum, where a large group of developers and NVIDIA engineers share their experience. Good luck on your project!
Hello, when I display the camera feed from my raspberry pi, i get some weird white dots that appear all over the screen. They seem to be permanent and not moving. Additionally, the image quality seems to be much worse than what your are getting out of your cameras. I tried buying the rpi camera from different suppliers, even from the one linked in your video description. Three different cameras in total, they ALL have this same issue and its very frustrating. Would you happen to know why or have any advice? Thanks!
I do not have any experience with your issue. Please ask this question on the official NVIDIA Jetson Nano forum, where a large group of developers and NVIDIA engineers share their experience.
Hi Jim, very nice video again! From my understanding this software synchronisation of the cameras is not perfect but close to perfect, right? Also shown in this video from Arducam: th-cam.com/video/MbLOcaAJ7Ug/w-d-xo.html Do you have an idea what the lag is (in ms, ns) between the cameras? Would the only option be to use global shutter and harware sync to get them perfect? Did you try to save the recorded frames (as idividual images) on the SD card or on a SSD and what would be the achievable frame rate in that case? I'm asking because I need to do postprocessing on the images and my script is so computational heavy that I cannot do it within the acquisition loop. Thanks Bart
In the video referenced, they use a global shutter and external trigger for the cameras. This is effectively a synchronized stereo image capture. It requires a different camera sensor, driver for the camera sensor, and external hardware trigger board. You describe a project which is beyond the scope of a quick answer. There should be enough information in this video to get you started in in the search to answer your questions. Thanks for watching!
Hi Bart, thanks for that question - I am also wondering about the maximum number of stored (HD) images per second, as my requirements look similar to yours.... BTW, I´m currently working on that issue on a StereoPi, which resulted in around 1 fps, however with some long waits of several seconds after 10 images or so. So in continuous burst mode not at 1 fps
hello. I am a student studying Jetson Nano and opencv for the first time. Because I use a translator, the writing may not be smooth. I want to save the video output from the code you uploaded as an mp4 file, but the VideoWriter function doesn't seem to work. Do you have any code to solve this? I've searched many examples, but haven't found a solution yet. Thank you for good material.
You are welcome. I do not have any code to share about video writing. You may ask on the official NVIDIA Jetson Nano forum, where a large group of developers and NVIDIA engineers share their experience. Good luck on your project!
Depends on the application. On a moving robot, you can aim the cameras in different directions, one being a front view, the second being a rear view. Same thing with side views. You can angle the cameras slightly from each other, giving you a wider field of view. You can point the cameras in the same direction a known distance apart and use the pair as a stereo camera. Using trigonometry, the stereo camera can estimate the depth of a given object from the cameras. On an inspection line, you can monitor two different areas on the line at once. Thanks for watching!
when i lunch th dual_camera_fps.py i have this message : Traceback (most recent call last): File "csi_camera.py", line 14, in class RepeatTimer(threading.Timer): TypeError: Error when calling the metaclass bases function() argument 1 must be code, not str
Hi (i'm new to embedded AI), would the xavier nx be able to take advantage of the 60 fps? also unrelated; for Jetpack 4.4 does opencv come with Cuda support? I cannot tell from the build info. Thanks so much for all your helpful videos!!!!!
Thank you for the video! I realized that the video output is much smoother with lower fps (e.g. 30) but there's so much lag for higher fps (e.g. 120). This seems counterintuitive to me... Shouldn't the video output have a faster/smoother output with higher fps?
Thank you for the long form video, walking through the code and explaining the caveats so clearly. We have missed you. Hopefully there will be more content coming our way soon. Cheers and stay safe.
This is a little departure, I usually try to keep the videos around 10 minutes. I'm glad you enjoyed this format. Thanks for watching!
Really nice work. Thank you for the teaching.
You are very welcome, and thank you for the kind words. Thanks for watching!
Brilliant video. New to the Nano. Gleaned ALOT from the vid. Never thought to thread the camera capture, was a "I could have had a V8." moment. Project I'm working on mandates high frame rates, now I got a clue. Big thanks.
Thank you for the kind words. Threading is usually the first trick to use on smaller machines if available. The goal is to try to take advantage of as much of the machine resources as possible. The smaller the machine, the more you have to know to make it performant. Thanks for watching!
@@JetsonHacks Yeah, concur regarding the need to get closer to the hardware & do a little "bit twiddling". I'm an "old school" coder. Cut my teeth on Univac main frames in the 70's-80's, transitioned to the 8086, 6800, & 68000 families in the late 80's, quite coding professionally in the middle to late 90's. Worked w/ the Nestor NI1000, back in the day (1,024 neurons in silicon) . My forte, language wise, was machine language, assembly (on several chips sets), C, C++, Ada, & several high level languages so Python is new to me, it being a relatively new OO language. So your "tutorials" are beneficial in multiple arenas. My dev machine is a HP DL580 (40 cores/80 threads, 1Tb memory, 4x 1080Ti's, etc.), am amazed @ just how much AI @ the edge can do (RPi 4 + NCS, Nano, AGX). Ever done anything combining the Nano + NCS (OpenCV + OpenVINO + CuDNN)? Think it would be interesting to encapsulate NLP "at the edge" versus the Alexa/Wolfram Alpha client server style implementation. Anyway, keep up the good teaching/experimentation, I'll be watching. What's the best way to subsidize your endeavors here on TH-cam, et al.? Feel the least I can do is throw you some pizza $$$. :)
I didnt order the Jetson Nano last year :) - but i came back to Your channel to check up if there's something new coming out :) - Thank You for a new video :)
You are welcome, and thanks for watching!
Thanks for your videos. I've learned so much by it.
You are welcome, and thanks for watching!
THIS IS AWESOME! thank you for this primer, next i need to feed each frame into a real time SLAM algorithm and build a volumetric representation of the space around the drone so it can A* path hmm
Thank you for the kind words, and thanks for watching!
where have you been? I am glad to see you best :)
I have been hiding in a room in my mind. Thank you for the concern, and thanks for watching!
Thanks a lot Jim! Awesome as always
You are welcome. Thank you for the kind words, and thanks for watching!
Thank you so much, your video gave me a great support!
You are welcome, and thanks for watching!
I am currently working on a project using an inland camera and when I try to test the camera this error shows up: "Error generated. /dvs/git/dirty/git-master_linux/multimedia/nvgstreamer/gst-nvarguscamera/gstnvarguscamerasrc.cpp, execute:557 No cameras available" can anyone help?
This example is for CSI cameras. I do not know what an inland camera actually might be.
Thank you for such a great educational video!:)
I'm glad you found it useful. Thanks for watching!
Yeeee you are alive
If you call this living ;) Thanks for watching!
Imo, firstly it's better to check if everything is ok with cameras using ls /dev/video* You can see 2 cameras with video0 and video1 names, means both must visible as devices. otherwise gst gives errors, wich are difficult to understand. Also dual camera sample didn't work for me with sensor_mode parameter, I've commented them out, then it was ok.
Thanks for the advice, and thanks for watching!
Hope you can get your hands on a Jetson Xavier NX when it's available! 😃
I hope so too! Thanks for watching!
Great manual, thank you. But could you cover case when second camera not in device tree. And how to fix that issue.
I do not know what that means.
I love your video. Could you make a video in which you use 02 CSI cameras and 02 Webcams? Thanks in advance.
Thank you for the kind words. Thank you for the suggestion, and thanks for watching!
I´ve purchased the IMX219-83 Stereo Camera from Waveshare but for some reason the tests don´t work... I would love to make that one work.
I am not familiar with that product. Please ask your question on the official NVIDIA Jetson Nano forum. Thanks for watching!
Hi! I purchased the cam too. For stereo vision with Jetson Nano, could you help me? I have tried to solve the connection between the cam and Jetson Nano.
i am currently using it, it works the same
maybe try changing CSI settings to IMX219 from jetson-io
(th-cam.com/video/eImDQ0PVu2Y/w-d-xo.html around 9-min guy uses the same program to configure pwm pins)
Awesome, my board has 2 CSI slots too,so does this mean we use these cheap pi cameras and create depth maps?
There are tradeoffs. Typically stereo cameras have global shutter and hardware sync which using two RPi cameras do not offer. The results won't be perfect, but you should be able to play around with the concept. Thanks for watching!
Hello Jim, i took a tutorial on your videos and did everything. But some point i got stuck with the syncing of two cameras in instrumened folder. I got an error like this:
python -u "/home/jetson/Documents/3D_workspace/CSI-Camera/instrumented/dual_camera_fps.py"
Traceback (most recent call last):
File "/home/jetson/Documents/3D_workspace/CSI-Camera/instrumented/dual_camera_fps.py", line 16, in
from csi_camera import CSI_Camera
File "/home/jetson/Documents/3D_workspace/CSI-Camera/instrumented/csi_camera.py", line 14, in
class RepeatTimer(threading.Timer):
TypeError: Error when calling the metaclass bases
function() argument 1 must be code, not str
Can you let me know what the error is..
Thanks in advance.
Hi ... thanks for the wonderful tutorials .. by the way, i just would like to know if you had experienced connecting flir lepton breakout board on jetson nano?
You are welcome. I have not used the Fair Lepton boards on the Nano. Thanks for watching!
Great video as always! I want to ask, How far can it detect objects from ?
Not sure
You have to use YOLO sample
Hi, I could only get a hold on a jetson nano A02 here in our country.. Is it possible to use 2 usb cameras in a stereo setup for depth estimation using A02?
I haven't tried it, but it should be possible. I would try two USB 2.0 cameras.
@@JetsonHacks thanks a lot for your suggestion, I've learned a lot from your contents, keep it up!
i am here to watch a tutorial but i watched too many advertisements. thanks.
You make it look so easy, but it's really not, at least for me. Looks like I'm going to have to devote a lot more time to this. There must be a way to do facial recognition from off axis. There's always a way to do something that seems imposible. I like that the fire extinguisher is always in frame. You should plug in your guitar and crank out some power cords at the end of some of your videos. I like saying rock on dude, instead of stay safe. I say that all the time.
Which version of opencv are you using
In the video, as stated in the JetsonHacksNano article in the description and referenced Github repository:
OpenCV 4.1.1
Thanks for watching!
Why not do a while loop around the cv2.waitKey checking the internal clock until 1000/fps has elapsed since you last presented a frame, and sleep for like, 1ms so it sleeps for exactly the right amount of time to match the frame rate?
Sounds like a nice extension!
Hi, Is there any way to turn off the CSI without turning off the jetson board. and then turn the CSI ON when required?
I do not know what "turn off CSI" means. You should ask this question on the official NVIDIA Jetson Nano forum. Thanks for watching!
Are there any differences in the board size between a02 and b01 ? Board screw holes etc? Tried finding the cad files for both, but didn't see them (still looking btw) :)
Hi there, wonderful video.. I have tried everything you said but still getting errors after running the Python code in visual studio.. not sure why.. I have installed everything, cython, numpy
I don't know what "getting errors" means.
@@JetsonHackswhen running the Python codes like face detect, simple camera scripts, it's throwing number of errors and I can't see the camera window. No matter what I do. Not sure if I am missing any plugins or software . I did whatever you have mentioned in your video
Thank you for this video
can you please provide the download like for the python-ide used
th-cam.com/video/2sHQBTtDz6c/w-d-xo.html
Do you know the specific model of you nano? Mine only has 1 CSI connector.
Jetson Nano B01 Development Kit. The earlier version with one CSI connector (A02) can be seen here: th-cam.com/video/dHvb225Pw1s/w-d-xo.html
hello Jim, does the 2 GB version work well with 2 cameras simultaneously without any problem?
The Jetson Nano 2GB only has one CSI camera port.
@@JetsonHacks thanks for the info.
Hi! Jetson nano B01 different with nano developer kit?
The Jetson Nano B01 is the development kit, second revision. Thanks for watching!
Hello, Can you please tell what is the maximum fps that can be achievable using Jetson nano and RPi camera(dual as well as single)? In general a RP camera specfi max is written as 120/90fps but I guess using OpenCV or bash command we can achieve higher. here using dual camera along with Jetson nano what's the max FRAME PER SEC limit is possible?
Please ask your question on the official NVIDIA jetson Nano forum, where a large group of developers and NVIDIA engineers share their experience. Good luck on your project!
Hello, when I display the camera feed from my raspberry pi, i get some weird white dots that appear all over the screen. They seem to be permanent and not moving. Additionally, the image quality seems to be much worse than what your are getting out of your cameras. I tried buying the rpi camera from different suppliers, even from the one linked in your video description. Three different cameras in total, they ALL have this same issue and its very frustrating. Would you happen to know why or have any advice? Thanks!
I do not have any experience with your issue. Please ask this question on the official NVIDIA Jetson Nano forum, where a large group of developers and NVIDIA engineers share their experience.
can jetson B01 run jetpack 4.3
The demonstration was shown on Jetpack 4.3. Thanks for watching!
Hi Jim, very nice video again!
From my understanding this software synchronisation of the cameras is not perfect but close to perfect, right? Also shown in this video from Arducam: th-cam.com/video/MbLOcaAJ7Ug/w-d-xo.html
Do you have an idea what the lag is (in ms, ns) between the cameras? Would the only option be to use global shutter and harware sync to get them perfect?
Did you try to save the recorded frames (as idividual images) on the SD card or on a SSD and what would be the achievable frame rate in that case? I'm asking because I need to do postprocessing on the images and my script is so computational heavy that I cannot do it within the acquisition loop.
Thanks
Bart
In the video referenced, they use a global shutter and external trigger for the cameras. This is effectively a synchronized stereo image capture. It requires a different camera sensor, driver for the camera sensor, and external hardware trigger board.
You describe a project which is beyond the scope of a quick answer. There should be enough information in this video to get you started in in the search to answer your questions. Thanks for watching!
Hi Bart,
thanks for that question - I am also wondering about the maximum number of stored (HD) images per second, as my requirements look similar to yours.... BTW, I´m currently working on that issue on a StereoPi, which resulted in around 1 fps, however with some long waits of several seconds after 10 images or so. So in continuous burst mode not at 1 fps
After a long time , Wow ...
Your GTC digital speech was amazing 👏
Thank you for the kind words. That speech was a lot of work, I'm glad some one watched it. Thanks for watching!
hello. I am a student studying Jetson Nano and opencv for the first time. Because I use a translator, the writing may not be smooth. I want to save the video output from the code you uploaded as an mp4 file, but the VideoWriter function doesn't seem to work. Do you have any code to solve this? I've searched many examples, but haven't found a solution yet. Thank you for good material.
You are welcome. I do not have any code to share about video writing. You may ask on the official NVIDIA Jetson Nano forum, where a large group of developers and NVIDIA engineers share their experience. Good luck on your project!
what's the benefit for installing two camera?
Depends on the application. On a moving robot, you can aim the cameras in different directions, one being a front view, the second being a rear view. Same thing with side views. You can angle the cameras slightly from each other, giving you a wider field of view. You can point the cameras in the same direction a known distance apart and use the pair as a stereo camera. Using trigonometry, the stereo camera can estimate the depth of a given object from the cameras. On an inspection line, you can monitor two different areas on the line at once. Thanks for watching!
when i lunch th dual_camera_fps.py i have this message :
Traceback (most recent call last):
File "csi_camera.py", line 14, in
class RepeatTimer(threading.Timer):
TypeError: Error when calling the metaclass bases
function() argument 1 must be code, not str
Hi (i'm new to embedded AI), would the xavier nx be able to take advantage of the 60 fps? also unrelated; for Jetpack 4.4 does opencv come with Cuda support? I cannot tell from the build info. Thanks so much for all your helpful videos!!!!!
Really helpful! Managed to up the FPS finally thank you.
You're welcome! Thanks for watching!
It is sad that Jetson does not support v1.3 rpi camera. :
Thank you for the video! I realized that the video output is much smoother with lower fps (e.g. 30) but there's so much lag for higher fps (e.g. 120).
This seems counterintuitive to me... Shouldn't the video output have a faster/smoother output with higher fps?
Not sure why you think that. You have to move 4x the pixels, and most monitors don't refresh that quickly (many are 30Hz to 60Hz).
Awesome! Can you do some object detection benchmarks next? :)
Thank you for the kind words, and thanks for watching!
Yey
Yey times two. Thanks for watching!
Hi!!!
Right back at you!
Very nice video. Do you know if RAW10 or RAW12 recording is possible?
Thank you for the kind words. I believe it is possible, but is highly dependent on which camera you are using. Thanks for watching!